Researchers have identified two failure modes in deep multi-agent reinforcement learning (MARL) applied to asynchronous pricing markets. These modes include tacit cartel formation among competing agents and actor-critic instability at high event rates. The study proposes a partial fix involving asynchrony and latency, which significantly reduces collusion but does not fully resolve the instability issues. AI
IMPACT Identifies critical failure modes in MARL for pricing, potentially impacting the robustness of AI agents in financial markets.
RANK_REASON This is a research paper detailing failure modes and a proposed fix for a specific AI technique.
Read on arXiv cs.MA (Multiagent) →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
